Content Repurposer
Turn one long-form piece you paste in (an article, transcript, webinar summary, or case study) into four channel-ready derivatives, an X thread, a LinkedIn post, a newsletter section, and a short-video script, each inside its channel's length and format limits, with every claim and number carried over exactly as written and any statistic the original does not source flagged instead of repeated as fact. Use when a finished piece has to go out on several channels and the derivatives must say only what the original says.
Paste an article, transcript, or case study and this skill returns four derivatives sized for their channels: an X thread, a LinkedIn post, a newsletter section, and a 45-to-60-second video script. It first builds a ledger of every claim and number in the source, draws only from that ledger, and flags any statistic the original never sourced.
Install to ~/.claude/skills/content-repurposer/SKILL.md
npx agentscamp add skills/content-repurposerRepurposing is where accurate long-form content turns into inaccurate short-form content: a number gets rounded, a hedge gets dropped, a "some customers" becomes "customers", and a stat that was already unsourced in the original travels to four more places. This skill treats the source as the only permitted material. It inventories every claim first, writes each derivative from that inventory, and then checks the derivatives back against it. It needs only pasted text, so it runs the same on claude.ai, in Claude Code, and in Claude Cowork; in a Claude Code project, /repurpose runs it on a file and writes the outputs to disk.
When to use this skill
- A blog post, report, or case study is published and the same idea needs to reach social, email, and video this week.
- A webinar or podcast transcript exists and you want the derivatives to quote what was actually said.
- Someone else wrote the piece and you need channel versions that do not drift from it.
- A piece is being reused months later and you need to know which of its numbers still have a source.
NOTE
The skill adds nothing to the source: no new examples, no rounded numbers, no stronger adjectives, no context it happens to know. If the original says "most of the teams we surveyed", the thread says that too, not "most teams". Hooks, calls to action, and closing lines the source lacks are written so they make no factual claim. Voice comes from a pasted brand-voice-profiler rules block if you supply one, otherwise from the source's own register.
Instructions
- Read the whole source and build the claims ledger. One row per factual claim, number, quotation, and named entity: the text as written, where it appears, and SOURCED (with the citation) or UNSOURCED. The ledger is the only material the derivatives may use.
- Pick the spine. One sentence the reader should remember, and three to five supporting points, each pointing at ledger rows. Every derivative is built from the same spine so the channels agree with each other.
- Write the X thread. Five to nine posts, each under 280 characters (or the limit the user states), with the character count printed after each post. The first post stands alone and makes a claim from the ledger, not a tease. Numbers appear exactly as in the source. The last post carries
[LINK]and at most two hashtags. - Write the LinkedIn post. 150 to 300 words. The first two lines carry the point, because feeds truncate the rest; short paragraphs; no more than one question; one call to action at the end. No engagement-bait openers.
- Write the newsletter section. A subhead, one paragraph of context in the second person, two or three bullets from the supporting points, and one
[LINK]. 120 to 200 words, ready to drop into an existing issue. - Write the short-video script. 45 to 60 seconds of speech, about 110 to 150 words, as a three-column table: time, spoken line, on-screen text. The hook lands inside the first three seconds and is a ledger claim, not a question. Mark any place a visual would need a number as
[SHOW: figure from ledger row N]. - Flag unsourced statistics. Any UNSOURCED number that appears in a derivative gets an inline
[UNSOURCED]marker and a row in a closing list, so the owner can source or cut it. Superlatives from the source ("the fastest", "the only") are flagged the same way. - Check the derivatives against the ledger. Compare every number, name, and quotation in the four outputs to the ledger. Nothing may be rounded, strengthened, or new. Fix drift before returning and note what was fixed.
- Assemble in this order: Claims ledger, Spine, X thread, LinkedIn post, Newsletter section, Video script, Flags, Checks applied.
Output
One Markdown document with the eight sections above. The four derivatives are ready to paste into their channels once the Flags list is cleared; the ledger stays with the document as the record of what the piece claims and what backs it. The video script can go to an editing tool such as Descript; the written derivatives can go to the content-editor agent for a voice and claims pass.
Example
Excerpt from repurposing a 1,800-word post about onboarding email timing:
### Claims ledger
| # | Claim | Location | Status |
| --- | --- | --- | --- |
| 3 | "teams that sent the second email within 24 hours saw higher activation" | para 4 | UNSOURCED (no study or internal data cited) |
| 4 | "we moved our own second email from day 3 to day 1" | para 6 | SOURCED (first-hand) |
### X thread
1/ We moved our second onboarding email from day 3 to day 1. Here is what changed and what did not. (96)
2/ The claim you will read everywhere: send it within 24 hours and activation goes up. [UNSOURCED] We could not find the study behind it. (134)
### Flags
- Ledger row 3 appears in thread post 2 and the newsletter bullet 1. Source it or cut it.The ledger discipline exists to prevent what the AI slop glossary entry describes; Claude skills for marketers shows where repurposing sits in the full set.
Related
- Brand Voice ProfilerDerive a reusable brand voice guide from five to ten writing samples you paste in: tone axes scored with quoted evidence, use and avoid vocabulary, sentence rhythm, formatting habits, dos and don'ts, three before/after rewrites, and a rules block ready to paste into a SKILL.md, with every rule traced to a sample and disagreements between samples flagged instead of averaged. Use when your team writes in a recognizable voice that nobody has written down, or the guide you have was never derived from what you actually publish.
- DescriptA video and podcast editor where you edit by editing the transcript, with Underlord, an AI co-editor, Studio Sound, AI voices, dubbing, and clip creation.
- RepurposeRun content-repurposer on a file or pasted text and write the X thread, LinkedIn post, newsletter section, and video script to content/repurposed/.
- Claude Skills for Marketers: 6 to Upload TodaySix Claude skills for marketers (brand voice, SEO brief, repurposing, landing page, email sequence, documents), how to upload them, and a prompt for each.
- Content EditorUse this agent to edit marketing and editorial drafts (blog posts, landing pages, emails, social copy, newsletters) for clarity and brand voice while marking every unsourced claim, statistic, and superlative with a [CITATION NEEDED] or [VERIFY] tag and a one-line reason, never inventing a source, and returning the edited draft with a change log. Examples — 'edit this post before it goes to the client', 'tighten this launch email and tell me which numbers I need to back up', 'check this page copy against brand-voice.md and mark anything we cannot prove'.
- AI SlopAI slop is low-effort, mass-produced AI-generated content — fluent, generic, and unchecked — flooding feeds, search results, and codebases.
- Anthropic's Marketing Plugin for Claude: Every Skill and Command ExplainedEvery skill in Anthropic's open-source marketing plugin for Claude Cowork and Claude Code, its connectors, the install commands, and how to fork it.
- Opus ClipAn AI clipping tool that turns long videos into scored, captioned short clips for Shorts, TikTok, and Reels, with a scheduler, brand templates, and an API.